How algorithms decide what we want or what we know
Decision-making by companies, governments and social networks increasingly involves more data and more sophisticated algorithms that are capable of detecting patterns that escape the human brain
When there is a hurricane warning, sales of flashlights go up at Walmart supermarket chain stores. They also sell more Pop-tarts, a type of stuffed cookie. When the company's analysts discovered it in 2004, every time the weather forecasters announced a storm, store managers would put out Pop-tarts at the front of the stores. Pop-tarts' sales exploded.
This mysterious correlation - why not cereal, chocolate or other types of cookies? - was discovered via two factors. The first: a historical series of transactions that recorded the items purchased by each customer, the total cost of the purchase, the time of day, whether there were sports competitions and even the weather. The second: an analysis of this massive amount of information by an algorithm that brought to light a relationship that had gone unnoticed to human eyes.
Data and more data
The information we give out without knowing it is processed by machines
This commercial success is just one example of what can be achieved with data, algorithms and computing power. "There has always been data, what happens is that now there is more than ever, and there will be more and more", says Josep Puyol-Gruart, a researcher at the CSIC's Artificial Intelligence Research Institute. Indeed, there are more and more connected devices and it is expected that with the so-called internet of things, when not only phones and computers are connected but also refrigerators and washing machines, even more data will be generated.
In fact, we generate data every day without even realizing it. When we turn the lights on and off, we give out information that power companies use to manage the electricity grid. When we turn on the water to take a shower, the same thing happens. When we buy with a credit card, when we use a search engine or interact on social networks, we generate even more information. These platforms collect a multitude of data: what we look at, at what time, how long we look at it, who we interact with, what we say, the speed at which we read, the speed at which we type, the speed of the internet connection and a long etcetera of data and more data that can be processed with increasingly powerful and sophisticated algorithms, which, in turn, have less and less human supervision. They are guided by a single goal: to predict what we will do or what we will like. "These algorithms are black boxes and this is the big problem", explains Gemma San Cornelio, researcher in digital culture at the UOC. "Even professional influencers find that content positioning algorithms change and they have to be constantly adapting to it", she points out.
Use or abuse
To improve covid lethality or to discriminate against riders, who deliver food to people's houses
The question, as always with any technological tool, from a knife to the most complex artificial intelligence algorithm, is the use (or abuse) that is made of it. As algorithms can detect patterns unnoticed by the human brain, they can be used, for example, as the researcher Carolina Garcia-Vidal has done, to reduce from 11.6% to 1.4% the mortality of patients admitted with covid-19 at the Hospital Clínic in Barcelona.
Financial institutions also use them. Based on historical data series and customer profiles, including postcode and nationality, algorithms calculate the risk of a loan and assign an interest rate that increases with the risk. This makes it more difficult for people with fewer resources to obtain a loan and costs them more money than those with more resources, who can pay it back more easily. And this, which seems so reasonable from the point of view of the interests of the financial institution, contributes to consolidate and increase inequality. Because the algorithm does not analyse the specific person asking for the loan, but what similar people have done until then. This lack of personalisation means that in some specific cases the algorithm's predictions are wrong. And the homogenization in decision-making, which saves human judgment, promotes vicious circles such as young people who are not hired for lack of experience or the long-term unemployed who cannot find work because they have been out of work for too long.
Algorithms that process large amounts of data are also used in some places to screen college students or workers, to predict where crime will occur most, or to set insurance rates. In most cases, as mathematician Cathy O'Neil explains in the book Weapons of Math Destruction, the lack of personalization of the process contributes to increasing inequality.
One of the latest examples that has made this clear is the riders' claim to know the algorithms used by delivery companies, something that will finally be possible soon thanks to the so-called rider law that the Spanish government is finalizing. These algorithms score the riders, and this determines their ability to choose schedules and orders. But those who can choose less often can't work, because the jobs are already assigned to them and this means that they have less and less points and therefore less and less opportunity to choose. Thanks to the new law, riders will at least be able to know what criteria the algorithm uses to award scores, so they will be able to decide how to work to get more points.
A business based on manipulating users' emotions
However, those who have taken the use of data and algorithms to another level are social networks and search engines. They are free and convey the idea that they are a window open to the world. Their business model, genuinely 21st century, is based on the constant capture of data and a unique algorithmic capacity. As has been said many times, if a digital product is free, the product is us, the users. In this case, the product is our growing attention, which is sold to advertisers with the guarantee, higher than ever, that the ads will be relevant to us. Therefore, the business of social networks is based on knowing users, knowing what will interest them and how they will react to certain emotions, to provide them with content that makes them spend more and more time connected so that they generate more data and can see more ads.
Google's former design ethics expert Tristan Harris calls it persuasive technology because these are applications that modify behavior. They are not just tools waiting for someone to use them, Harris says, but platforms that use human psychology to generate profit. Evolution has shaped us over millions of years to be social animals, and we place importance on relationships and what people think of us. One of the mechanisms that time and chance have selected to regulate these relationships is the generation of dopamine in the brain, which is produced during positive interactions and feeds the pleasure centers. Social networks exploit this mechanism as no one has ever done before. If in the elevator or at a red light we can't help but look at our cell phone, it's because it's hardwired into our biology.
The key to the success of these platforms is the ability to provide relevant content to users. These contents and interactions are selected or promoted for each user at each moment based on all the information that has been accumulated, including their mood. And this can create a certain addiction and generate discomfort and frustration, especially in teenagers, an age group in which some networks encourage the enthronement of aesthetic canons that are often based on unrealistic filters, also selected by algorithms. "Social networks are likely to generate these problems, but I have never seen anyone with addiction problems", explains psychologist Cristina Martínez. From a psychological point of view, "the consequences of abusive use of social networks are much less than those of addiction to video games or gambling", she says.
Networks only promote contact with like-minded people
The most defining characteristic of social networks is the creation of so-called bubbles o resonance chambers. To keep people connected, algorithms encourage contact with like-minded people, so that the promise of an open window to the world easily becomes a porthole into the user's own world. "If you don't have an active attitude, you're left with a very restricted view of the world", explains San Cornelio. "Either you search for information consciously or the information comes to you filtered by algorithms", says Puyol-Gruart. "The problem is that there is no transparency", says Karina Gibert, professor and director of the Intelligent Data Science and Artificial Intelligence centre at the UPC. "They don't explain how they choose the content they show you and you can't free yourself from this choice", she adds.
This dynamic favors polarization, especially in political aspects, which acts as a very effective retention mechanism. So does the proliferation of false information, which feeds back into this polarization and spreads six times faster than the truth. Much has been said about a possible artificial intelligence that could destroy humanity in the future. But perhaps not so much has been said about the artificial intelligence used by social networks that has already brought about real changes in the world. In 2016 Donald Trump spent $44 million on nearly 6 million Facebook ads, during a campaign that contributed to a measurable and studied polarization of American society (Hillary Clinton posted 66,000). The same network was used in 2016 in Burma to whip up violence towards the Rohingyas, a Muslim minority living on the country's west coast.
According to Gibert, one of the aspects that make the situation more complex is that "what is considered ethical depends on the cultural context". A group of experts from the European Commission drew up recommendations a year ago for using artificial intelligence algorithms in a way that respects data and privacy but, according to the professor, "they are at odds with what is ethical in other countries". In addition, she adds, "the market for digital applications is geographically delocalized, and all of this creates a conflict of competitiveness for Europe".
Regulating an algorithmic world
Citizen culture and regulation with a demand for transparency
Data will not disappear. Neither will computing power or algorithms. But algorithms are not only built with data, but also with human decisions about what data is used and how. "From the technical side, there is little we can do", says Puyol-Gruart, "regulation has to come from politics". In Gibert's opinion, one of the keys to improving the current situation is "for there to be a culture of citizenship and for everyone to have criteria for deciding how they relate to technology, what data they give up and under what conditions".
On the other hand, "companies would have to be transparent and explain with what criteria they design algorithms and with what data they train them, which is perfectly compatible with industrial secrecy", explains Gibert. As the regulation of all this algorithmic power is complex and, therefore, distant, the actual use or abuse of this technology depends on each individual. So if you want to open up to the world and not feed your digital bubble, it is better not to read this text if it comes to you through a social network.